模型预测控制神经网络快速求解优化问题的最新进展  

Recent Advances in Fast Solving Optimization Problems Using Neural Network for Model Predictive Control

作　　者：李星辰 李天勋 游科友[1] LI Xingchen;LI Tianxun;YOU Keyou(Department of Automation,Tsinghua University,Beijing 100084,China)

机构地区：[1]清华大学自动化系,北京100084

出　　处：《信息与控制》2025年第1期66-78,共13页Information and Control

基　　金：科技创新2030—“新一代人工智能”重大项目(2022ZD0116700);国家自然科学基金项目(62033006,62325305)。

摘　　要：模型预测控制(MPC)是一种具有稳定性保证的最优控制算法,但需要在线求解优化问题,这极大限制了其在嵌入式系统中的部署。得益于深度学习和计算硬件的发展,离线训练、在线推理的神经网络近似MPC方法受到了广泛关注,在实际控制系统中具有优异的表现。本文讨论了两个关键问题的最新进展:神经网络的结构设计、数据集生成和训练方法;神经网络近似误差的应对方法,以保证闭环系统稳定性和满足约束。最后,给出了神经网络在GPU(图形处理器)和FPGA(现场可编程门阵列)上的快速部署案例,展示了使用神经网络近似MPC的实际性能优势。Model predictive control(MPC)is an optimal control algorithm with stability guarantees,but it requires solving optimization problems online,which greatly limits its deployment in embedded systems.Thanks to the development of deep learning and computational hardware,neural network approximation MPC method with offline training and online inference has received widespread attention and has shown excellent performance in actual control systems.This paper discusses the latest progress on two key issues:1)the design of neural network structures,dataset generation,and training methods;2)methods for dealing with neural network approximation errors to ensure closed-loop system stability and constraint satisfaction.Finally,we present case studies of fast deployment of neural networks on GPU and FPGA,demonstrating the practical performance advantages of using neural networks to approximate MPC.

关 键 词：模型预测控制 神经网络 约束满足 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP13[自动化与计算机技术—控制科学与工程]